文献类型：期刊文献

中文题名：结合自编码器和卷积神经网络的推荐系统

英文题名：Recommendation system combining auto-encoder with convolutional neural network

作者：刘钟涛[1];刘兰淇[1]

第一作者：刘钟涛

机构：[1]河南财经政法大学现代教育技术中心,河南郑州450046

第一机构：河南财经政法大学

年份：2021

卷号：42

期号：2

起止页码：559-566

中文期刊名：计算机工程与设计

外文期刊名：Computer Engineering and Design

收录：CSTPCD;;北大核心:【北大核心2020】；

基金：河南科技攻关基金项目(18210221022)。

语种：中文

中文关键词：推荐系统;降噪自编码器;观点识别;情感识别;卷积神经网络;评分预测

外文关键词：recommendation system;denoising auto encoder;opinion recognition;sentiment recognition;convolutional neural network;rating prediction

摘要：传统推荐系统以评分作为推荐依据,没有分析与利用用户的评论内容,导致推荐系统存在推荐准确性低和数据稀疏性的问题。针对这种情况,结合降噪自编码器和卷积神经网络提出一种推荐系统。通过卷积神经网络学习评论内容在方面级的情感和观点,基于降噪自编码器对方面级的观点集合进行归纳和分组,以三阶张量分解技术为基础,推断出用户对项目的综合评分。实验结果表明,该系统有效提高了推荐系统的推荐性能,优化了推荐项目的顺序。
Classic recommendation systems usually treat the ratings as the recommendation basis,they do not analyze and take advantage of the comment content of users,which lead to problems of low recommendation accuracy and data sparsity of recommendation systems.In view of this,a recommendation system combining auto-encoder with convolutional neural network was proposed.Aspect level sentiments and opinions of comment contents were learned through convolutional neural network,the aspect level opinions were summed up and grouped based on denoising auto encoder,and the overall rating was inferred based on three order tensor decomposition technique.Experimental results show that the proposed system improves the recommendation performance of recommendation systems effectively,it also optimizes the ranking of recommended items.